The Butler
Back to briefings
GitHub / Security workflow

Turning on GitHub AI Scan is not the same as running a scan

· The Butler · 4 minute read

A metallic blue-eyed robot Butler reviews documents at a desk, wearing a tuxedo and top hat.
Butler editorial illustration, not a product screenshot.

A security feature can be enabled while no one has actually used it. GitHub's November 13 update makes that distinction unusually important: organizations can now manage AI Scan for repositories through the REST API, but enabling the setting does not launch a scan.

The announcement adds a repository-scan control at organization and repository level. This is useful for a consistent rollout. It is not evidence that the code has been inspected, that findings have been reviewed, or that a fix works.

For a busy maintainer, the worthwhile change is the ability to separate those jobs and give each one an owner.

What the new switch controls

GitHub says an organization can enable repository scanning while allowing individual repositories to opt out. If the organization disables it, a repository cannot enable it independently. Enterprise restrictions still apply. The feature remains a public preview and is not available on GitHub Enterprise Server. Each scan consumes Code Security AI credits; turning on the setting is not a free-running background inspection service.

The REST documentation is the place to check permissions and the current request format before automating a rollout. Read the stored setting back after changing it. A successful configuration request is a useful audit record, but name it accurately: configuration applied, not repository secured.

Start with one repository and a budget

GitHub's AI Scan overview describes an on-demand scan of the default branch, complementing rather than replacing CodeQL. Availability depends on repository ownership, visibility and the relevant security entitlement. Findings enter the code-scanning workflow, where people still need to investigate them.

That suggests a modest first deployment. Choose a repository someone actively maintains, confirm it is eligible, identify who can initiate the scan and decide how much usage is acceptable. Avoid a broad enable-everything script as the first experiment. It can produce a long list of changed settings without establishing whether anyone can act on the results.

Write the success criterion before starting: a completed scan, an identified reviewer and a recorded decision on each finding worth pursuing. Counting enabled repositories is easier, but it measures coverage of a switch rather than progress on security.

Keep four receipts

First, keep the configuration record: which repository was enabled and under whose authority. Second, keep the scan record: when it ran and which code revision it covered. Third, record triage: the findings accepted for investigation, dismissed with reasons or assigned for remediation. Fourth, record verification after a change.

Those receipts can be short. A small project does not need a heavyweight dashboard to avoid confusing these stages. A shared issue with links and dates is often enough to show whether work stopped after setup.

Be particularly careful with the last stage. GitHub documents that repository scans are not automatically rerun after a fix. Scheduling another review is part of the job, not an optional flourish. A proposed fix and a checked fix should remain different labels.

Automate the repetition, not the conclusion

Once the first repository has gone through the whole cycle, automation becomes more valuable. You know the permissions required, where the reports appear, who owns them and what a useful follow-up looks like. Expand from that working example instead of treating the API as a reason to skip it.

Our practical learning paths take the same small-project approach: learn one complete workflow before increasing its reach.

The Butler's recommendation is simple: use the new API to make configuration repeatable, then measure completed scans and reviewed findings separately. A green setting is the beginning of the process, not its conclusion.